Data Analytics
I believe that most problems that have yet to be solved are just data problems. We can understand almost everything with the right data at hand, and that is reflected in my projects.
The Numbers
Demo
Model statistics
Explore Claude and Jev calls over time. Tokens combine input and output. A snapshot of real usage from the Fantasy Football Agent, 09-28 to 10-27.
Daily trend
By model
View exact model statistics
The Problem
All of the websites and projects that I build have as many analytics and measurements as I can get away with. The demo above shows why.
There was a bug in my weather forecast engine, and when it failed, it failed quietly. No alerts went off and nothing broke. But the result was that it was calling Claude an absurd number of times, which could have led to huge API costs!
Thankfully, a gate already written into the code did not allow any tokens to be used from an engine that had failed. Still, calling Claude that many times was not ideal. Because of this dashboard and the data it reported, I was able to catch it and fix it!